Taxonomies and activities for organisational effectiveness

A wonderful harmony is created when we join together the seemingly unconnected.

(Heraclitus, quoted in von Oech, 2001)

It’s time to move beyond a consideration of infrastructure (and taxonomies as part of infrastructure) to the role that taxonomy work can play in the basic activities of organisations as they pursue their goals. We will be guided in this task by another taxonomy of sorts, Donald Marchand’s strategic information alignment (SIA) framework (see Figure 4.1 ).

Figure 4.1.

Marchand’s strategic information alignment framework

Source: Marchand (2000).

Marchand developed this framework to help organisations align their information management practices with their business objectives, so it is based on a simple taxonomy of four categories of business focus. In its original intended use, organisations assess how effectively they are using information to support each category.

The first and the most primal business focus (yes, this framework has an implicit hierarchy) is the management of risk. Risk comes from both internal sources as well as the external environment. After risk, organisations must worry about reducing costs – or, more accurately, increasing the productivity of their resources and assets in relation to their costs. This dimension is about margin and efficiency. Once they have set their internal house in order, organisations tend to focus on their income streams – outwardly, to their customers and markets, where they must add value in order to compete. They must understand customers (information gathering) as well as inform and educate them, so this involves a two-way information flow. The most mature focus is on innovation. If risk, costs and customers are all well served, then resources turn towards the creation of new reality (Marchand, 2000: 24–8).

Marchand’s framework is useful because it is oriented towards information work and matches information needs to the primary types of activity that contribute towards effectiveness. Two important activities (for us) that he does not cover, strategic planning and talent management, can be accommodated when we look more closely at knowledge management applications of taxonomy work in our next chapter.

Risk

Categorisation is, of course, fundamental to the management of risk. Different kinds of risk must be identified and grouped together based on origin, severity or remedy. Risk intelligence systems need to identify the signals or clues that would indicate particular categories of risk and put in place monitoring mechanisms (strategic early warning systems) so that these signals are picked up whenever a risk is emerging (Gilad, 2001). Control, avoidance or remedial strategies need to be prepared for each category of risk.

In increasingly complex and uncertain environments, new and unexpected risks emerge with unsettling frequency. Here our established taxonomies of risk are of little help. The greatest contribution the organisation can make is in early category creation.

When unusual events occur of which we have no prior experience, this is the same as saying that they do not fit into our established categories for understanding and dealing with things. But it is a basic human coping mechanism to seek to place the unusual event into our taxonomies of response. This is because we use taxonomies to identify unknown things. Taxonomies are sense-making frameworks to the extent that they allow us to take an unknown thing and relate it to other similar things that we do know in our taxonomy. If it has feathers, wings and a beak, we know it’s a bird, even if we don’t know what type. This is closely related to the practice of diagnosis.

If we categorise an unknown thing ineptly, our response strategies may be inappropriate. When the disease SARS first launched itself into the world via the Hotel Metropole in Hong Kong, it looked – and was treated – like a type of viral pneumonia. In such diseases, the lungs gradually fill with fluid until the victim can no longer breathe. Standard procedure when this happens is to perform a tracheotomy – cut a hole into the victim’s windpipe and allow the fluids to drain from the lungs.

What was unnerving about SARS was that it didn’t behave ‘normally’ – i.e. according to the category of disease to which it had been assigned. Healthcare workers who were treating SARS patients started to get sick and die in alarming numbers. The strength of the categorisation actually hindered appropriate response – it couldn’t be viral transmission the experts said, because this type of virus can’t survive for very long outside the body.

The countries with the most highly developed healthcare systems often fared worse than those without. The first SARS infected country to come off the World Health Organisation’s danger list was Vietnam. I’m told that when a senior Vietnamese official was asked how they had managed to control the SARS infection rates so effectively, he replied:

We saw what was happening in other countries, so we did two things. First, we put troops round the hospitals – people could go in, but they couldn’t come out. Second, we stopped doing tracheotomies – which the French had taught us as standard procedure. All that fluid coming out under pressure from the lungs had to be the transmission vector. We just pumped the patients full of antibiotics and hoped for the best.

Whether this anecdote is true or not, tracheotomies were highly risky procedures in the absence of very strict controls against infection via droplets, airborne transmission or direct contact (Ohara, 2004).

Now, of course, we know that the SARS corona virus does not behave like other respiratory tract viruses, and it does remain active and infectious for considerable periods outside the body. We now have a new category for it.

The SARS case illustrates one of the dangers of strong taxonomies. Human beings are dangerously prone to making first bets on scanty evidence – ‘but it looks like viral pneumonia, it must behave the same way!’ And once we are sure it’s in the correct taxonomic slot, we’ll ignore overwhelming evidence to the contrary.

HIV AIDS shows another aspect of the importance of category creation. When it was first noticed at the beginning of the 1980s, it was a deeply mysterious disease. It behaved like nothing else known to medical science. It was initially thought to be a skin cancer disease since that was how it became visible to doctors – the puzzling thing was this was previously an extremely rare condition limited to narrow racial groups. Gay men seemed particularly at risk, and so it was labelled a ‘gay disease’. Its easy transmission through unprotected sex or reuse of needles by drug addicts made sure its early categorisation was not a scientific one at all. It entered our taxonomies of morality and blame faster than it entered our scientific taxonomies of treatment. For years, it was socially visible but invisible to taxonomies of treatment.

However crude or discriminatory the categorisations of those early years, and however much they slowed the establishment’s efforts in addressing the disease, they still had some effect. Haemophiliacs also got the disease, so it could also be transmitted through the blood. Gay men and haemophiliacs getting sick in other ways began to reveal the many other symptomatic expressions of the disease and helped identify it as an immune disorder. Finally, other groups of sufferers also became visible.

The challenge for the medical community in respect to HIV AIDS was that they had to learn about it from zero. Nobody knows how long it took before it was even ‘seen’ as a disease, since nobody knows how many people’s deaths before the 1980s were AIDS-related but classified on death certificates as due to other causes (death certificates must use the International Classification of Diseases) (Bowker and Star, 1999: 124–5). Frozen tissue samples from 1968 suggest that it was killing people at least since the 1960s (Osmond, 2003). If it doesn’t exist in the taxonomy it is invisible. If it is invisible it is unknown. If it is unknown it is untreatable.

It’s hard to study something that doesn’t exist in a taxonomy. The virus itself was isolated in 1983 but not definitively linked with AIDS until the following year. It was not given a place in the International Committee on the Taxonomy of Viruses classification until 1986 (Osmond, 2003). Why did it take so long? Because it behaved like nothing else in our taxonomies of disease – we had no related categories to compare it with so that we could learn about it by proxy.

Indeed, the discipline of virology was just starting to emerge as a separate discipline in the 1980s – previously, the study of viruses was scattered across the taxonomy of research disciplines according to the virus hosts – plants, animals, human beings. So there was little reliable integrated knowledge of viruses at all. And viruses are notoriously resistant to classification by traditional scientific means. They mutate in their hosts, they cause wildly different effects, they trade genetic material with their hosts. In short, they refuse to remain in a neat biological box. There are still biologists who believe that viruses do not constitute a true taxonomic category at all because different viruses originate in different species and belong to those species (Bowker and Star, 1999: 96–8). Of course, if they are scattered across the taxonomy of living things, it is virtually impossible to study them in any depth. Gathering things together into a taxonomic class is the only true way to study, compare and learn. Taxonomies determine by inclusion and creation of categories what you can see and know, but also by exclusion or scattering what you cannot see or know.

Now, of course, by creating a medical category and building a body of questions and knowledge around that category we know considerably more about HIV AIDS and other viral diseases, and by extension have strategies and drugs to deal with other viral and auto-immune diseases.

Taxonomies are incredibly powerful for the recognition of risk, but taxonomy flexibility and new category creation are vitally important for the anticipation and management of new risk. Our only warning is that strong taxonomies can also inhibit the recognition of new risk. Flexibility is an important requirement.

Case study 4.1. Department of Homeland Security digital library.

In 2002, in the wake of the 9/11 terrorist attacks, the United States government decided to formally recognise a new risk on a new scale and established the Department of Homeland Security. Soon afterwards, the Department’s Office of Domestic Preparedness saw the need for a digital library that could be made available to a broad range of local, state and federal policy-makers.

This library would contain the latest information on homeland security strategy, policy and current research so as to ensure common awareness, consistent response and regulatory compliance. It would cover a broad swathe of risk areas that had been identified for the United States: border security, disaster management and response, epidemiology, inter-governmental relations, intelligence, law enforcement, money laundering, weapons and technology. In short, it would form part of the United States government’s risk awareness and risk response framework.

Information would need to be distributed in a way that supported consistent and coordinated awareness and response across many different agencies. Moreover the content itself was collected from a wide range of agencies in different formats and styles of presentation. Officials very quickly identified the need for good metadata and categorisation to broker a common search across this diversity and enable effective and relevant retrievals from the system.

The original intention was to use auto-categorisation, where a programme indexes the materials automatically, and assigns categories based on the content of documents and defined semantic rules. However, this was difficult to apply in practice because the topics covered by the collection spanned many different disciplines, the terminology and format of documents varied widely, and the area of homeland security itself was a new creation, with fluid and evolving terminology and categories of its own.

The user community would need a common language of search spanning different disciplines and working contexts. It was decided that a taxonomy would have to be built from scratch, and only then could this be used to teach the auto-categorisation engine how to recognise the contents of documents and tag them in ways that diverse user communities would recognise in browsing and searching.

The taxonomy was built over two years, manually. It was built inductively from the content for the digital library as it was collected. This is a classic ‘cluster and sort’ technique in taxonomy building. Because it was such a new domain and contained many cross-disciplinary linkages, a large body of content was necessary to be able to define the overall shape and scope of the domain, discriminate the boundaries of useful clusters of content and label them appropriately. At the same time, the vocabulary of homeland security was itself emerging and changing, and required some time to stabilise into a common, widely understood vocabulary.

The taxonomy development process was emergent and iterative, mirroring the domain of homeland security itself. The first stage was a relatively simple tree structure taxonomy of topics and sub-topics. This was then expanded into a thesaurus by adding alternate terms and cross-relationships between topics in the taxonomy structure. Finally, the taxonomy was turned into a faceted scheme by adding schedules for geographic region and event type. It now had three facets: Topic, Geography and Events. A fourth facet covering Agencies was under development in late 2005.

In 2004 the taxonomy team acquired Teragram auto-categorisation software and settled down to train it. There are several different techniques for training auto-categorisation software – you can write semantic rules for it by specifying the semantic patterns within documents that indicate particular topics; you can give it training sets to analyse of typical documents for each topic; you can let the tool crawl your content and make suggestions and then correct the mistakes. Teragram uses a semantic rules-based approach.

Defining rules is a very labour-intensive, skilled process. The topics in the taxonomy must be described to accommodate the variety of ways that they are expressed in the documents themselves – another reason for having a good-sized representative collection of content to build from. Hard work though it is, the advantage of using this approach is that as topics, vocabularies and issues continue to change, the rules themselves can be edited and adjusted promptly and with relatively little effort. The rules for the taxonomy term ‘Hurricanes’ were revised in the wake of Hurricane Katrina, for example, to include rules that would capture documents describing levees and flooding.

The Homeland Security Digital Library project is a good exemplar of how taxonomies are critical to enabling coordination around risk. In this case the project was founded on the need to make a common resource base available to the various agencies involved in homeland security – and quickly evolved into the need to establish a common language and set of categories for that community. It also illustrates some of the difficulties of taxonomy development in new, cross-disciplinary, emergent or rapidly changing knowledge domains – and strategies that can be deployed to deal with those difficulties:

• ■ Collect a lot of content.
• ■Work iteratively, from simplicity to depth.
• ■Wait for the vocabulary to stabilise.
• ■Define topics against their context of use.
• ■Use facets to expand your collection’s openness to different perspect
• ■Design for flexibility and openness to change.

Source: Pitts (2005).

Costs

Managing costs usually requires an intimate knowledge of your internal processes and constant information flows to monitor and control them. In this way you can maximise the productivity of your resources, minimise waste and reduce mistakes, redundant effort and re-work.

The most obvious taxonomic work in support of this is systems mapping – or, more precisely, process mapping. Of course, not all process maps are taxonomies (because they don’t always fulfil our three requirements for a taxonomy), but they are first cousins and have many family resemblances. Business process maps translate very easily into taxonomies.

Moreover the mapping process itself often uses taxonomic frameworks and establishes common vocabularies – as taxonomies do – to ensure that the mapping project team members are all viewing and describing the processes in the same way. For example, in Six Sigma process improvement methodology, teams develop an early, high-level process map called a SIPOC – an acronym for the categories to be analysed: Suppliers, Inputs, Process, Outputs, Customers (see Figure 4.2 ) (Rath & Strong, 2000: 11).

Figure 4.2.

SIPOC process map

This activity bounds the process to be mapped and analysed, and establishes a common vocabulary for all the key elements in play. Exactly the same framework (under different names) can be used in knowledge mapping exercises to take a knowledge and information view of a business activity (see Figure 4.3 ). Here, for each business activity, you establish and name the key knowledge and information inputs (and their suppliers) and outputs (and their customers), which can then be used as a basis for:

• ■ creating an information and knowledge inventory;
• ■making resource decisions;
• ■identifying gaps, redundancies and duplications;
• ■mapping coordination and collaboration opportunities;
• ■improving information flows;
• ■informing an ‘Activity’ facet in a taxonomy;
• ■informing a records management facet in a taxonomy.

Figure 4.3.

SIPOC knowledge mapping template

In many respects, the process of mapping is as important as the product. Mapping exercises are best done with all the process’s stakeholders and actors present in a room with a large whiteboard and lots of sticky notes. You ask the people present to map out the process in question, building out the detail into micro-steps indicating decision points and actions from the high-level map composed for the SIPOC exercise. Most important, they are asked to map the process as is – that is to say, not what the manual says the process should be, but how it actually works in the real world, warts and all. The reason for doing this as a collective exercise is that the exercise will uncover variations in understanding and variations in practice among the different actors. It will also uncover variations in how the processes are described – ripe ground for misunderstandings and rich material for taxonomy building later on. Many of these steps will have documents and information resources attached to them.

The mapping exercise stabilises a common standardised view of the current state among all the players, and it is this baseline view which can be analysed for loops, errors, dead-ends, and circumnavigations ripe for abbreviation.

If you have never done business process mapping, you will be surprised by the degree of variation in practice and understanding – among different practitioners as well as between practitioners and the official manual. Sometimes, where different grades of staff are involved, this variation may not come through in a group setting, so you will need to compare across several mapping interviews. The real illumination always comes when the whole group sees the variances, bad and good, and collectively resolves them.

In one consultancy I undertook for a non-profit organisation that also conducted commercial training programmes, the training unit manager was convinced that the front-office staff – who served the whole organisation – knew how to handle potential customer enquiries. According to the procedure, calls should be put through to him or his deputy immediately. He was convinced because it was a simple process, he had trained them himself and he was regularly receiving telephone calls as per plan.

Mapping interviews with the front-office staff unearthed at least three other pathways. Two of them were new hires and didn’t know that the deputy manager could also receive enquiry calls so she was rarely disturbed. If the manager was out (which was frequently, because he also trained) they improvised – either forwarding the call to one of the trainers or attempting to answer the questions from the brochures to hand. Sometimes they took messages, but these were limited to name and number with no details of the enquiry and were passed to the department secretary. It might take some time before the unit manager got them.

Improvisation in the face of organisational dislocations or lack of specific guidance is one of the biggest causes of variation. It’s not always a bad thing. When my local bank was acquired by another bank some years ago, it temporarily suspended international remittance services while the two operations were being integrated. Only the acquiring bank provided these services. As a customer, I didn’t discover this until I tried to use the service. My teller didn’t blink – she picked up the phone, called her colleague in the other bank (in a branch just across the road), gave her my details and asked her to prepare the documentation for me while I crossed the road. Impressed, I asked her if this was their standard procedure. She said, ‘No, we are not really supposed to do this, but we arranged this privately because a lot of customers were getting upset.’ This improvisation was an excellent indicator for a coordination opportunity that had been missed, but should really have been built into the standard process and scaled. Officially, it didn’t exist.

As this example shows, collective or individual mapping interviews are not always enough. You have to go and see the process in action to uncover the issues that the players may be unaware of or think unimportant. In my non-profit organisation sometimes, it turned out, front-line staff made category mistakes – they interpreted the call as meant for a different department, resulting in customer frustration and multiple call backs. Sometimes, while trying to be helpful, they gave the wrong information based on outdated information, or mis-understanding, or partial information. Sometimes the call was put through without checking whether the manager was at his desk and it was left to ring until the potential customer gave up.

In my non-profit organisation the collectively constructed map allowed all the players in the process to identify improvement opportunities which included:

• ■ a simple set of diagnostics to identify the enquiry type and customer type (two implicit taxonomies here!);
• ■standard information scripts for general enquiries, together with a call record log where calls were categorised by topic;
• ■checking that the manager or deputy were in before putting through calls;
• ■message forms containing more customer information and details of the enquiry for non-standard enquiries.

The discovery method for taxonomy building is very similar to the discovery method for business process mapping. This should not be surprising, since all but the most mechanised of processes rely on the supply, movement and application of information and knowledge. Taxonomy development fulfils the same stabilisation and standardisation role as process mapping, and is frequently an extension of process mapping. Taxonomies meant to support the internal activities of an organisation must do some form of process mapping.

Like process mapping, taxonomy work enhances visibility, control and improvement. It is based on current language in use and current real-world practice. The collective mapping approach allows for improvements that are mutually agreed, prioritised and owned by the actors involved, giving them significantly greater chances of adoption. Conversely there are few process improvement projects that do not rely on establishing new information flows and semantic standardisation to measure, monitor and assure consistency.

While mapping work and semantic standardisation is an important dimension of cost management in organisations, it’s not the only way that taxonomies can have an impact. Taxonomy work is also fundamental to the management of resources and the planning work around resource deployment.

Let’s take a planning taxonomy from the domain of agriculture as an example. Crop rotation is the discipline of changing the nature of crops grown on any given land every year to ensure its sustainable fertility. It is based on a simple classification of crops according to their interactions with the soil – the nutrients they take out, and the nutrients they leave behind. This forms the basis for a crop rotation plan, just as the BCG matrix forms the basis for a product development plan (see Figure 4.4 ). It’s one of the most basic things to know as a farmer if you want to minimise your artificial fertilisation interventions and farm the same land sustainably into the future.

Figure 4.4.

Crop rotation plan for a kitchen garden

Planning and resource management always requires some form of organisation of resources according to a taxonomy, whether it is implicit or explicit.

Taxonomic strictness can vary of course. Companies tend to go through cycles of growth and capacity building, when taxonomic rigour is relaxed and cost management is placed second to opportunity building. They acquire based on opportunity, not exact fit with the traditional categories of their business. The other side of that cycle is focus, when simplification is the order of the day, and cost control and cost effectiveness become paramount. Here, the contribution of taxonomy work becomes much more apparent, because its primary purpose is to consolidate – i.e. reduce to agreed categories. After all, one of the primary goals of taxonomy work is to simplify – either access to, or management of, a knowledge domain.

Case study 4.2. Unilever’s brand simplification exercise.

In early 2000, Unilever announced that it was slashing its 1,600 portfolio of brands to the 400 ‘jewels’ that already earned 90 per cent of its revenue.

It was a conscious exercise in simplification. In a June 2003 presentation in Chicago, Unilever executive John Ripley quoted a McKinsey report: ‘A lot of our clients have been choking in their own complexity, resulting in little innovation or employee fulfillment’ (Ripley, 2003).

The more brand labels you have to manipulate, resource and manage, the harder it is to navigate your landscape of brands and products effectively. Unilever expected their simplification strategy to double growth (optimistically, as it turned out), but also to save some $3 billion in four years, half from business process simplification and half from procurement simplification. In procurement for laundry products alone, they would eliminate more than two-thirds of Unilever’s 152 sets of specifications for perfume ingredients.

When you look at the final stable of products that emerged from their simplification exercise, mapped into a tree structure in Figure 4.5 for the purposes of clarity, it’s immediately obvious that their brand simplification exercise was also, in all respects except in name, a taxonomy exercise.

Figure 4.5.

Unilever’s taxonomy of brands

Unilever’s brand management structure falls easily into a taxonomy tree structure: divided first into three industries, then into 13 categories, with the 400 or so brands behind those.

Like many other real-world taxonomies, it has some untidy features and polyhierarchical links inside it. Some brands appear in multiple categories (e.g. Dove, Slim.Fast, Lipton). Others have a family of sub-brands beneath them at the fourth layer (e.g. Surf, Heart), while others move immediately to a single product or cluster of products (e.g. Snuggle, Hellmann’s).

Many brands are in transition – Carb Options is an aggregating meta-brand for the health conscious and is being stamped onto other Unilever brands (such as Hellmann’s, Lipton and Skippy) in an attempt to give them a lateral (polyhierarchical) association. In 2004, Unilever adopted a new corporate mission centred on the theme ‘Vitality’. This has now had a knock-on effect, as a selection of brands across the taxonomy become identified as ‘Vitality’ brands, creating new polyhierarchical linkages. As complexity grows after their simplification, the tree structure will become increasingly difficult to manage and will likely transform into a faceted structure in the longer term.

This exercise illustrates how powerfully taxonomies maintain the organisation of resources, people and products in support of strategic or operational effectiveness, just as much as they can help organise knowledge and information content for retrieval. It also shows how simplification and complication coexist in a constant dynamic. Above all it should demonstrate that there is plenty of work for taxonomists beyond information retrieval.

Customers and markets

Customers are mysterious and difficult creatures, largely because, despite our best efforts, they insist on making their own decisions. And they change. Like viruses, they mutate and evolve, stubbornly evading predictable categorisation. They are a taxonomist’s nightmare.

We might build taxonomies organised around our own internal worlds – for example, taxonomies of sales enquiry types for sales teams, or taxonomies of complaint types for call centres. Japanese cosmetics company KAO has a centralised customer service department to handle complaints, but it also has a policy that complaints are dealt with as close to the root of the complaint as quickly as possible. Their customer service staff have to categorise and despatch complaints to 150 different destinations as soon as they come in. That requires a very good taxonomy of complaints mapped to the organisation structure, and good, consistent, diagnostic mechanisms to ensure the complaint ends up in the most customer-friendly place (Marchand, 2000: 43).

But most of our customer taxonomies are determined less by our world and more by theirs. I have just watched a TV advertisement for Knorr packet soup (coincidentally, a survivor of the Unilever brand simplification exercise). We see a husband opening a packet of yummy tomato soup and preparing it while his tired wife unwinds after a hard day’s work. The narrator describes the events in a tongue-in-cheek recipe format ‘Allow wife to steam gently in bath’. The catchphrase, as our loving couple snuggle up with spicy tomato soup on the dangerously white sofa, is ‘Knorr: Recipe for Living’. The ad is witty, heartwarming and relevant. It’s not just about soup and its nutritional value any more – it’s about stories and contexts. And those stories and contexts are determined by things that matter to us – or at least, what marketeers think matter to us.

Knowing what matters to our customers and being able to provide it is the key to successful competition. As futurist Rolf Jensen puts it:

The English AGA stove has had no product development since 1922. The price is in the $10,000 to $15,000 range, installation not included. Yearly sales are at 7,000 units, but the stove has a clear future because of nonexistent product development. Time magazine quotes marketing manager Ian Heath as saying, ‘What other kitchen appliance can promise family togetherness? They (the customers) want family life and AGA is at the center of that wish. They see the AGA in a very emotional sense.’ This is not a status appliance, although the price tag would seem to indicate otherwise. It is a lifestyle appliance with a story of family togetherness, something that abounded in the old days, back when family values and gender roles were not up for discussion and before mass production robbed products of their spirit. (Jensen, 1999: 39)

How do we get to know what our customers think? In a large, noisy and confusing marketplace, how do we even know who our customers are? Which ones matter most? Or who we want to be our customers? Which of the opinions we unearth matter more, because they represent a whole category of customers and not just an individual voice? How do we know what a customer will respond to, and buy, even though they are not yet aware of this incipient desire? In a world of conglomerates, global markets, mass media and the Internet, where our customer population is so diverse and distant, how do we learn, organise and deliver – quickly? The answer is, at least partially, in taxonomy work. The role of taxonomies, after all, includes domain simplification, description and charting for reliable and speedy navigation.

In the management of risk the taxonomic traits that matter are category recognition and knowing when it’s time to create new categories; in the management of cost the taxonomic traits that matter are standardised description, collaborative mapping and simplification. When it comes to customers, the taxonomic trick, the real competitive edge in adding value for customers, is in discovering categories that work. Our taxonomies of customers must reflect their worlds, not ours. The onus on category discovery is very high.

Customer segmentation is the traditional approach – and in its way is quite Aristotelian, insofar as it attempts to categorise on objective, observable attributes. Depending on your goals, you can segment by intrinsic properties or by behaviours – or both. If you want to know who buys what, you’ll segment on demographic data like age, gender, marital status, as well as attributes of geography, socio-economic position, lifestyle. If you want to know why they buy, you’ll segment on frequency of purchase, consumption patterns, price sensitivity, media exposure and so on. Add to this the formulas for customer lifetime value and cost of acquisition and you start to get categories that look actionable (Fleisher and Bensoussan, 2003: 162–79).

The Moor Hall Health Club in the UK felt that the so-called ‘grey market’ of seniors was under-exploited in a largely youth-oriented industry. In their area, seniors made up 38 per cent of the population. Against all expectations, the club was remarkably successful, doing better even than clubs that targeted the younger population.

A segmentation analysis of this market using a matrix taxonomy, comparing the three dimensions of financial assets, social support needs (loneliness factor) and whether seniors’ motivation for staying fit was intrinsically or extrinsically driven, generated a typology of eight potential customer profiles. Moor Hall had effectively targeted facilities, events and marketing strategies towards the ones that made business sense – resulting in above average performance for a health club and a trailblazing model for health clubs in a greying population (Fleisher and Bensoussan, 2003: 177–8).

Moor Hall had some advantages: they were situated in a local community which they could get to know well and which had a fairly well defined population. What if we could do segmentation analysis on a very large scale? This is the stuff of which myths are made, particularly among those who sell data warehouses and OLAP systems. In a data warehouse you gather all of the segmentation data compiled from surveys, research and transaction records from multiple systems. OLAP stands for online analytical processing and is a data manipulation technique which allows you to examine your vast reservoir of data by any combination of attributes, e.g. ‘show me the pre-Christmas spending patterns of married bald men over forty in Germany’. You can slice and dice your data in myriad ways, and what you are doing is looking for that combination of winning attributes – the winning categories in your million different taxonomic combinations – that will give you actionable customer profiles. It’s essentially a matrix taxonomy with unlimited dimensions of comparison.

This is the stuff from which the fabled beer and nappies story comes from – which I have never been able to find an authoritative source for. The story goes that transaction data analysis applied in a supermarket chain discovered that nappies and beer had especially high sales on Friday evenings – when sliced and diced to perfection, it was discovered that husbands did the nappy shopping for the wife just before the weekend and were compensating for the shame by stocking up on beers for the weekend football on TV – hey presto, nappies and beer were thenceforth placed close to each other, just in case the husband was tempted to forget the nappies (upsetting most senses of proper categorisation along the way). The supermarket chain decided that the category of football-and-beer husbands with young children was one worth having.

There are also lots of more credible stories about business intelligence applications for customer segmentation and categorisation based on behaviours and demographic data (Vitt et al., 2002). But the approach is far from problem free.

The attributes you are using and comparing are not as objective nor as stable as they look. Data warehouses gather data from many, many different sources, and the data they collect has been input in many, many different contexts with varying standards and definitions for the attributes that are assigned. Remember that Vivian Alvarez had eleven variants of her name in the various Australian immigration department databases. And that was just her name. What about attributes that are more open to interpretation?

What, for example, puts someone in the category ‘unemployed’? Is it somebody who is between jobs? Between graduation and first job? If I’m self-employed and don’t currently have a job, am I unemployed? Do I have to be on social security? How long do I have to be unemployed before I’m unemployed? Your wife will probably have a different view on this compared to your local social security office. Such distinctions are important: one nameless government department took a conservative interpretation of the category ‘unemployed’ because they wanted to send a political message about their success in reducing unemployment. Then they realised that they were funded based on how many people were ‘unemployed’ and their interpretation became more liberal. Needless to say, long-term trend analysis was somewhat complicated by this switch. Historians might later become curious about why there was a sudden leap in unemployment in the year of the categorisation change.

Don’t be fooled: all categorisation is subjective and fluid. The issue for taxonomy developers is: how much variance can you permit consistent with your goals and how can you improve consistency where you need it?

Different data collection processes may be gathering segmentation data inconsistently. In France I might check the ‘35–45’ age range box, while in Singapore I check the ‘40–50’ age range box. How can the segmentation data be compared?

The truth is, data warehouse and OLAP systems have enormous data normalisation problems arising from inconsistent standards and variant understandings of the meanings of the categories being used. It’s rarely easy to figure out and correct for all these variances without going back to the contexts of data collection and original categorisation decisions. They are a model lesson for taxonomy developers: in the usefulness of scope notes to give context to the interpretation and selection of taxonomy terms, and in the importance of training and standardisation if the purpose of your taxonomy will be compromised by lack of precision.

For this reason, despite the alluring specificity of the data warehouse stories, segmentation is a crude instrument unless allied with a more intimate knowledge of your customers. It only gives very broad profiles and very general patterns, and really only works well as a very high-level pattern-identification technique. As Fleisher and Bensoussan point out, ‘predictions based on customer characteristics may fail to materialize in sales due to the impact of more influential behavioral factors that were not identified during the segmentation analysis’ (Fleisher and Bensoussan, 2003: 172).

It certainly does not get us inside the head of the customer with sufficient granularity to understand their dreams, desires, stories and resonant contexts. The increasing trend towards product and service mass customisation, personalisation and co-creation increases the demand to find better ways of discovering actionable customer categories. In fact, it turns out that the famous Moor Hall story was not a pure customer segmentation analysis story at all – the segmentation analysis was done retrospectively and then used to explain the decisions they had already made intuitively based on their local knowledge.

If we are not confident about how to categorise our customers then the logical next step is to allow them to categorise themselves. The trouble here is that we have to give them something to select. One of the ways we do this – usually badly – is in automated voicemail menu systems. Here we construct ‘if-then’ decision trees – essentially a taxonomy of choices, guiding our callers into the taxonomy of ways that we are organised to deal with them.

Voicemail has a huge disadvantage as a tree structure. Visually, we can navigate 12–15 categories in a single list, and we can take time to interpret the less obvious terms in the list before we make our choice. Aurally, we are easily confused above four categories in a list, and we have to rely on our memory to compare the choices and figure out which category belongs to us or promises something in the right direction (Byrne, 2001).

Limiting the breadth of the top level of a tree means that the taxonomy has to be deeper – which means that the ‘real’ topics will be pushed deeper under more intermediate generalities, multiplying the number of choices we have to make to get to them. Remember (from Chapter 2) that cognitively human beings start in the middle of a taxonomy with base level categories somewhere between extreme specificity and extreme generality. We know what we want – we may not necessarily know how this company sorts it and generalises it.

No wonder the aural taxonomies of voicemail have such a reputation for being confusing and frustrating. The generic topics at the top are too generic to be meaningful, and we may not discover our error until we get three or four levels down and then find we have to start again. Unfortunately, the people who design these trees break one of the primary rules of taxonomy building (use the language and categories of your audience), because they almost invariably build the choices around the organisation’s categories and not the customer’s categories. The study of usability may have become respectable in website design but has not yet penetrated telephone voicemail. (It’s improving only now because voice user interface design for computers is developing fast.) In 1996 a Reuters study found that voicemail menu navigation was taking an average of 20 per cent of the call time (Reuters, 1996). Things don’t seem much better a decade later.

About a year ago I spent some time in London and rented a house from friends who were travelling. I was taking over the telephone bills in my name, but wanted to keep the old number for my friends’ return. I called the British Telecom helpline and chose the option ‘Customer Service’ and then ‘Moving House’. However, the options here were requesting the telephone number of my old house in the UK which, being resident in Singapore, I didn’t have. Not being able to navigate backwards (no option for that) I called again, but couldn’t find any other category at the top level that matched what I wanted. Finally I pressed zero to speak to an operator (a long wait), who put me through to sales. The sales person wasn’t able to deal with my enquiry and passed me back to someone in the depths of customer service who passed me to somebody else. I wasn’t able to find out which category I should have chosen or ended up in because the person I was speaking to didn’t know. I was saved only by social networks among the telephone support staff. I got what I wanted in the end, but I’m sure there must have been a way through the voicemail tree. This teaches us a few further lessons about taxonomies:

• ■We should never underestimate the capacity of human beings to misunderstand our labels.
• ■We can never assume our staff will know what our categories mean any more than our customers will.
• ■Aural taxonomies are particularly prone to errors because of the lack of immediate ‘checking’ context – designers need to anticipate mistakes and provide navigation out of them.
• ■Social networks compensate for taxonomy mistakes, but the more effective they are, the more they conceal taxonomy faults.

There are more positive developments. In recent years, work on the development of personas and archetypes has gone some way towards developing categories that more truly reflect our customers and our customers’ world (Pruitt and Rudin, 2003; Goodwin, 2001). We are getting closer to rich category discovery from our audiences, and the tools for doing so come from anthropology.

Case study 4.3. Club Med, storytelling and archetypes.

In the 1990s, Club Med discovered that it was getting a lot of customer complaints. Over 40 per cent of customer dissatisfaction was linked to miscategorisation – customers were choosing or being recommended the wrong type of resort for their needs. For example, adventurous singles would end up in resorts oriented towards families with kids, sports enthusiasts in party-oriented resorts, party-goers on lush, deserted islands, and so on.

When they looked closely at their customer profiles, they identified the following persona-types among their customers:

• ■Tubes – family types oriented towards comfort.
• ■Celebrators – young people oriented towards partying.
• ■Epicureans – oriented towards luxury and comfort.
• ■Cultivated – independently minded, oriented towards exploring the local culture.
• ■Activists – oriented towards adventure and sport.

They used this small taxonomy as the basis of their ‘Key to Happiness’ programme, which is a self-completed diagnostic questionnaire available to prospective customers in all Club Med retail outlets. It is designed to identify the characteristics that match customers to their closest persona. The resorts themselves are styled according to the preferences of each category. The taxonomy was being used to allow customers to self-select into resorts that were most appropriate to their preferences (Horovitz, 2000: 63).

Building customer personas in this way is often no more than imaginatively extended customer segmentation. The original segmentation analysis can be enriched through focus groups and customer observation. By marrying this with customer self-categorisation, Club Med had created a categorisation process that was much less likely to result in category mistakes.

But these personas are still a long way from getting us inside customer’s heads. We cannot guarantee that they are uncontaminated by our own internal categories and concerns. Dave Snowden of the Cognitive Edge network (formerly Cynefin Centre), building on work done in the 1990s with the IBM Institute for Knowledge Management, has developed interesting new techniques for generating personas (and the implicit taxonomies that lie behind them) that come directly out of customer self-perceptions.

The key to the technique is the use of storytelling. Instead of bringing customers in for focus groups and leading them with questions structured by our internal concerns, you bring them in to tell stories – about their experience with your product, your brand or the particular lifestyle area you are interested in. The stories are written down as they are told. The same customers then work with their stories to identify stereotype characters appearing in their stories and the attributes associated with them. In a two-stage process drawn from cultural anthropology techniques, the attributes of these stereotype characters are removed from the original stories and clustered according to how customers see the relationships between them. Each cluster becomes the basis for a new fictional character that expresses all those attributes. These characters are called archetypes, because they are grounded in the collective experience of that customer group and express general sets of values, behaviours and attitudes across the group (Snowden, 2000a). Archetypes are extremely interesting persona-sets from a marketing standpoint, because they are drawn directly from the customer experience by customers themselves. The biases of corporate category systems have not been imposed on them.

Collections of archetypes derived in this way form a special kind of taxonomy called a typology. They are different from the matrix typologies we discussed earlier in Chapter 2 and the segmentation typologies discussed earlier in this chapter because they are arbitrary clusters of attributes – they have not been constructed along set dimensions of comparison. They basically exist as a list of characters. But they do represent subliminal types of personality or ‘customer identities’ that exist within the perceptions of the customer community that has produced them, and have very high resonance when ‘played back’ to those customers. They fulfil the ‘mapping’ criterion for taxonomies if the process has covered a sufficiently representative cross section of your target population. The full archetype set represents a map of the cultural perceptions of your population in regard to your target topic – lifestyle perceptions, customer experience, life aspirations.

These archetypes, and the story-collections that underpin them, form a powerful resource for understanding the category worlds of customers. They can be used, as in the Club Med example, to drive self-categorisation initiatives that attract customers towards categories that are natural to them. They can be used to organise products and services in ways that make more sense to customers. They can be used as proxy customer types to interrogate for value adding ideas or to test innovations.

Innovation

Taxonomies can be as dangerous for innovators as they are for risk managers. Complexity guru Dave Snowden is fond of saying at conferences, ‘If you want to focus on the future, don’t start with a taxonomy project, because a taxonomy represents your past.’ After all, if innovation is about getting out of the box, taxonomy work is mostly about putting things into boxes. Strong taxonomies can blind managers to innovation opportunities just as much as to risk, because true innovation, by definition, is as yet uncategorised. Extreme risk and extreme opportunity are both extreme because they come at us unforeseen and unrecognised.

There’s a noteworthy degree of truth in this, but taxonomy work can also support innovation in two major ways:

• ■Innovation can sometimes be stimulated by taking existing categories and breaking them apart or combining them in unusual ways – we see this in lateral leaps of the imagination where new connections between categories are made that are not naturally resident in the taxonomy as it stands.
• ■Existing taxonomy categories can provide a foundation for disciplined ways of exploring innovation opportunities – they give a structured foundation for exploring the unstructured. We see this in technology mapping activities of research and development teams.

In both cases, a taxonomy can form a lever against which innovation work can be done, by using the taxonomy against itself. Innovation almost always puts stress on a taxonomy, but the taxonomy itself is a tool in that process, a stable counterbalance against which innovation work is defined. Furthermore, innovations must always end up being incorporated into our taxonomies, because without this integration into our infrastructure and basic operations, we cannot make our innovations productive.

Innovation’s relationship to taxonomy work is therefore strained: taxonomy work enables innovation, but innovation challenges taxonomies, and taxonomies must grow to accommodate innovations.

The relationship is not entirely dialectical, however. The taxonomy work we have labelled category discovery together with the use of taxonomies for sense-making are two aspects of taxonomy work that are particularly supportive of innovation work.

Category-busting activity is a classic innovation generation strategy – but to do this, you need a firm base of categories to challenge or recombine. Without the old categories as reference points it’s hard to recognise and place the innovation when it arrives. One of the five key attributes of innovation success, according to Everett Rogers, is compatibility – by which he means that the innovation can be matched to existing categories and understood sufficiently to support adoption, even if this is by comparison or contrast (Rogers, 1995: 224–34).

Cirque du Soleil is a classic example of innovation by category-busting. It advertises itself as circus, but circus reinvented. Despite its title, it lacks many of the traditional features of circus – the ring format, the animals, the big name acts. ‘It is not quite a circus and not quite opera or theater either, but takes elements from them all’ (Williamson, 2002: 3). In some respects, it differs from all three – in the anonymity of its performers, for example. It is successful not simply because it breaks traditional categories, but by self-consciously displaying its category-busting as innovation. The resonances of circus, street performance, theatre, opera and ballet are all superbly put on show to be appreciated as a sophisticated blend.

R&D departments in large corporates must make a discipline of continually breaking down categories and making new blends. Many use science and technology mapping techniques that embody clear features of taxonomy work. Science and technology mapping relies on the assumption that primary scientific research holds the key to downstream commercial innovations, so it tries to identify indicators of promising categories of research. Mappers track investments in particular areas of technology by industry players and research scientists to identify promising new technologies; they map the citations in scientific primary research papers, to look for clusters of disciplines interacting with each other in novel ways, and they map patents, categorised by the basic science areas behind them. Such activities indicate promising combinations of scientific categories – sometimes in unpredictable areas.

In the late 1990s a global consumer products company had determined to make a significant innovation in its soap products. They would do this by moving beyond the traditional knowledge domains in soap research: dermatology, physical chemistry and clinical medicine. They created basic science maps of research being invested in by their competitors as well as in the public research sector, across the whole science universe. They were looking primarily for ‘high-performing’ research areas – areas that promised commercially viable technologies, determined partly by the intensity of citations and size of the research clusters, partly by competitor investment in those areas, partly by the nature of the research itself.

Once they had identified promising areas, they mapped these in greater detail, looking for patterns or weak linkages with dermatology that might show promise. Their mapping activity picked up interesting work being done by a group of geophysicists who were working on mathematical modelling of the cracking of the earth’s surface in times of drought. ‘Cracking’ was also a category of interest in dermatology, so they contacted the geophysicists and initiated a collaboration. They were eventually able to incorporate the geophysical modelling of cracking into a new soap product that had superior anti-cracking capabilities (Pauker and Whitaker, 2000: 21).

This entire project simultaneously relied on established categories and strong taxonomies (to be able to map the interaction of scientific disciplines), and undermined them (by looking for opportunities in weak linkages across widely separated categories). And it used the taxonomic work of semantic mapping to be able to achieve the job.

Case study 4.4. Unilever Research and disposable taxonomies.

Innovators don’t always like rigid taxonomies, but they usually appreciate taxonomy work. Here’s how Unilever’s former head of knowledge management Adrian Dale described the approach of research teams at Unilever in 2001.

Unilever operates in a diverse range of fast-moving consumer goods markets. Its products span ice cream to tomato sauces and shampoo to washing powder. Its scientists work in diverse fields from anthropology to mathematics with every physical and life science in between. Building durable company taxonomies across these ranges has been impossible and will probably always be so. Instead, Unilever Research has taken a pragmatic ‘just-in-time’ approach – creating taxonomies quickly and dynamically for a specific task and then throwing them away. To do this, mind mapping and creativity tools have been deployed in facilitated sessions to develop a shared language for the worlds in which research teams operate. (Dale, 2001)

For the research manager seeking innovations, it appears a waste of time to focus on big taxonomies representing all of the disciplines at play. He or she won’t bother with taxonomies of their own discipline, because they already know the structure of their discipline backwards. They don’t want to know the deep dark recesses of every other discipline, either. They want to be able to pull together an interdisciplinary domain map of the relevant areas in the relevant disciplines at a sufficiently high level to be able to make investment decisions. The research teams want to be able to pull together topic oriented maps at a more granular, very specific level, as they work on issues within the research programme.

The approach to solve this problem is not external expert-driven taxonomy development at all. The members of the team build their instant taxonomies themselves, using whiteboards and sticky notes.

Each mapping session lasts 2–3 hours and involves dialogue and arguments between the participants. In this process a shared context and language is developed, building on the combined knowledge and experience of the team. Between each session, time is required for the facilitators to tidy the documentation and for participants to reflect on their experience. The number of mapping sessions required depends upon the scale of the field but a minimum of 3 is recommended. (Dale, 2001)

The results of the sessions are transferred into durable documents using mind-mapping software and are used in:

• ■ patenting – to identify patent infringements and opportunities: the maps indicate the science areas that need to be checked for possible prior art;
• ■ environment scanning – to programme search agents to crawl external repositories for intelligence in the identified fields;
• ■ intranet – to structure specialist intranet sites.

Dale is obviously aware in his account that his approach may not sit easily with traditional taxonomists.

We cannot claim any rigor for the taxonomies we build using this methodology and make no apology for this. They are fit for purpose, just-in-time and throw away by design. However, their leverage is enormous. We are easily able to engage teams in the exercise and get them thinking in great depth about their fields. (Dale, 2001)

His account gives three important insights about good taxonomy development practice, however. First, the flexible, needs-driven taxonomy activity he describes is a great model for how to build any taxonomy:

• ■Know why you are doing it.
• ■Involve the taxonomy users.
• ■Negotiate a common understanding.
• ■Have a clear idea of how it will be implemented.
• ■Don’t get too attached to it.

Second, it demonstrates that, as Jean Graef puts it in her commentary on this case, taxonomy work goes far beyond the stereotyped view of a tree structure and a thesaurus for a content management system: ‘The bottom line is that taxonomies can take different forms depending on the applications they are designed to support’ (Dale, 2001).

Third, despite Dale’s protests, his account assumes a strong taxonomy infrastructure in the background. Interdisciplinary maps of the depth and complexity he describes cannot be drawn out of thin air. Every specialist in the room during a mapping session is bringing detailed, strong, public taxonomies to bear. It’s just that they are not on paper. They are in their heads.

Unilever has cropped up a lot in this chapter. They appeared in our discussions of managing costs, customers and innovation. Taxonomies are supposed to promote serendipitous discovery, and it so happens that serendipity brought three useful, documented examples from Unilever which were hard to refuse. For that we have to thank them, because taxonomy work, in common with other infrastructure work, is not widely documented and not easily made visible. The contexts and the issues this organisation faces, together with our other examples, are generic enough. The basic taxonomy work of categorisation, category creation, semantic standardisation, collaborative mapping, domain simplification, category discovery and category mixing pervades the basic work of organisations.

This should not be taken as a landgrab attempt for taxonomists, but an invitation to see the value of conscious taxonomy disciplines and approaches for all the core aspects of what makes an organisation effective. In Chapter 3 we described how important taxonomy work was within the information infrastructure of organisations, and we made the point that without strong infrastructure, coordination on a large scale becomes difficult and failure prone. In this chapter we traced the many ways in which taxonomy work either consciously or unconsciously informs our basic business activities. Taken together, it becomes very clear that taxonomy work holds a wider range of application and use than simply as a tool for information retrieval.